Professional Documents
Culture Documents
Where g is the link function through which probability of default (p, equivalent to probability of
default) is related to the explanatory variable X, p is the probability of event, prob(Y=1) is the
parameter vector, and X is the matrix of explanatory variables. The expected probability of
default can be analytically solved for p through logit transformation where
Where is the logistic distribution function. (.) A
The logistic regression specified above assumes a natural logarithmic relationship between the
explanatory variables and the ratio of event (foreclosure) to non-event (no foreclosure),
otherwise known as the odds ratio. Thus the natural log of the odds is linearly related to the
explanatory variables.
This form is desirable in that the logit function, g(p), is linear in its parameters, and the logistic
transformation always generates an outcome p probability between 0 and 1. Estimates of the
parameters are obtained by maximizing the sample log likelihood function:
where L(.) is the likelihood function, yi=1 for a defaulted loan and 0 for a nondefaulted loan, and
p is the probability measure.
SAS and other statistical software contain statistical procedures to perform logistic regression.
The SAS procedure LOGISTIC is used to estimate the parameters of the regression model with
options to generate goodness of fit and other diagnostics designed to assess model robustness.
1 i=
( )
( ) |
|
|
X
X
X Y prob p
exp 1
exp
) ( ) 1 (
+
= A = = =
| X p p p g = = )) 1 /( ln( ) (
U.S. Prime RMBS Loan Loss Model Criteria 30
August 15, 2011
Structured Finance
Univariate analysis is initially performed for each variable individually to understand
distributions, strength, and forms of relationships to default before performing the analysis in a
multivariate context. Each variable is assessed for the need of additional transformations
based on the nature of relationship to default. Nonlinear transformations are applied when
necessary to improve the fit of the regression. Performance of the regression was measured by
nonparametric statistics for rank ordering/separation ability (i.e. Somers D, Kolmogorov-
Smirnov, or K-S). The tests are performed on all 2.2 million data and on the traditional loan
sample only in line with target portfolios. All tests showed the model had high power in rank
ordering and separation ability. The main focus in measuring the models performance is the
point-estimate accuracy on benchmarked vintage projected default rates where average rates
of predicted versus observed foreclosures were measured and compared for loans originated
from 1991 through 2007. The model successfully captured the dramatic transition in credit risk
for the peak default vintages of 2005, 2006, and 2007, as seen in the chart below for traditional
loan sample.
0
20
40
60
80
100
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
0
2
4
6
8
10
12
No. of Loans Tr adi t i onal PD Act ual and Pr oj ect ed New Model
Model Perf ormance: Act ual vs. Predi ct ed Tradi t i onal Sampl e
(000) (%)
0
50
100
150
200
250
300
1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
0. 00
5. 00
10. 00
15. 00
20. 00
25. 00
No. of Loans Al l Sampl e PD Act ual and Pr oj ect ed New Model
Model Perf ormance: Act ual vs. Predi ct ed Al l Sampl e
(000) (%)
U.S. Prime RMBS Loan Loss Model Criteria 31
August 15, 2011
Structured Finance
Appendix D: Economic Risk Factors
The impact of economic factors on future defaults and losses is captured by the National Risk
Index (NRI) and regional risk multipliers provided by UFA, which are indicators of home price
forecasts and a number of other econometric measures. UFAs analysis is used to derive the
quarterly UFA Mortgage Report risk multipliers on a state, zip code, and national level. The
UFA multipliers are incorporated into Fitchs new PD model and raise or lower the expected
default probability on a given mortgage loan to reflect national and regional economic risk
forecasts.
State and zip code-level risk multipliers represent the level of expected risk over the life of a
loan relative to the national average on a constant quality basis. For example, if the UFA
default multiplier for a state is 0.90, expected defaults in that state are 90% of those for the
average loan in the U.S.
Supplementing the state-level risk multipliers with zip code multipliers provides increased
granularity in the default and loss risk analysis, since conditions at the regional level can exhibit
more disparate trends than those indicated at the state level due to variations in industry
concentration/diversification, employment growth, personal income, demographics, and other
factors. Both national and regional components are applied to the PD, aligning Fitchs base
case and stressed expected loss for newly originated loans with prevailing and forecast
economic conditions.
National Risk Index
The NRI provides a default forecast for loans originated today relative to loans underwritten
during the 1990s. Thus, the 19902000 economic climate serves as a benchmark from which
todays metrics are measured and is the comparative basis for future defaults. UFA assumes
the quality of the borrower, loan characteristics, and legal environment remain constant; thus,
changes in the index reflect only changes in current macroeconomic conditions.
Each quarter, UFA evaluates economic conditions in the U.S. and assesses how those
conditions will impact future defaults, prepayments, loss recoveries, and loan values for
nonprime loans. The NRI reflects changes in economic measures, such as real GDP growth,
real consumer spending, business spending, national unemployment rates, CPI inflation rates,
mortgage rates, national house price appreciation, and housing permits.
Because UFA updates the index quarterly, credit enhancement levels for pools with similar
characteristics could vary from quarter to quarter. To reflect the risk of further economic
deterioration and to provide for more stable credit enhancement and ratings through the
economic cycle, stressed index value floors are applied to all investment-grade rating
categories. The stressed level will be evaluated with each quarterly update from UFA. Should
the NRI rise more than expected, the stressed multiplier may be increased accordingly. If and
when the NRI begins to decrease, Fitch will re-evaluate the stress and may adjust it
accordingly. Any changes to the stressed level will be announced in the quarterly Fitch
research, which will be available at www.fitchratings.com.
Regional Risk Multipliers
UFAs analysis of regional risk takes into account state and local economic metrics, such as
personal income and distribution, employment growth, housing construction, and other
U.S. Prime RMBS Loan Loss Model Criteria 32
August 15, 2011
Structured Finance
indicators. It also factors in a demographic component, which includes unemployment rates
and population growth, as well as a political component that considers local taxes and zoning
regulations. The resulting multipliers reflect the areas expected lifetime default risk in
comparison to the national average. For example, a multiplier of 1.10 indicates that loans
originated in this region have a 10% higher default risk than the national average in the 1990s.
The regional risk multipliers are updated quarterly and are available at www.fitchratings.com.
Variability in credit enhancement levels could occur if there are significant changes in regional
concentrations among pools with similar credit characteristics. Also, unlike the NRI, where one
value is applied to all loans, risk multipliers vary by location; thus, regions with rising risk
multipliers could offset those declining in risk.
0. 0
0. 5
1. 0
1. 5
2. 0
2. 5
1990 1992 1994 1996 1998 2000 2002 2004 2006 2008 2010
Hi st ori cal Nat i onal Ri sk Index
Sour ce: Uni ver si t y Fi nanci al Associ at es, LLC .
U.S. Prime RMBS Loan Loss Model Criteria 33
August 15, 2011
Structured Finance
Appendix E: Heat Maps for Sustainable Home Price Model (2006 and Current)
Overvaluation in 2011 as Predicted by Sustainable Home Price Model
At or near sustainable
10-20%
20-30%
30-40%
40%+
Overvaluation in 2006 as Predicted by Sustainable Home Price Model
At or near sustainable
10-20%
20-30%
30-40%
40%+
U.S. Prime RMBS Loan Loss Model Criteria 34
August 15, 2011
Structured Finance
Appendix F: Summary of Changes to Model Exposure Draft
Model Component Exposure Draft Final Version
Regression Analysis
1. Roll Rate Methodology
Default projections for recent vintage collateral were based on
delinquency roll rates, defaults and prepayment performance over the
past two years to extrapolate future defaults. Fitch assumed the behavior
would continue at the same rate for the remainder of the life of the pools.
Fitch's default projections were updated to consider actual performance through
April 2011. The agency's projections for recent 'peak' vintages are based on five
year prior experience to capture some pre-crisis performance, reflecting Fitch
expectation of future stabilization in default performance.
Frequency Of Foreclosure Variables
1. Sustainable Loan to Value Ratio (SLTV)
SLTV was not an explicit PD variable in the agency's exposure draft.
The agency's view on current sustainable borrower equity was caputured
by two separate PD variables - Sustainable Market Value Decline and
Original Combined Loan-to-Value Ratio.
The Sustainable Loan to Value Ratio (SLTV) was included in the agency's final PD
regression model. This variable effectively combines sMVD and OCLTV to provide
a view on borrower sustainable equity, which is the most explanatory variable in the
agency's new PD regression model.
2. Cash-out Refinances 68% PD penalty relative to purchase loan baseline.
PD penalty increased to 100% relative to purchase loan baseline based on new
regression results.
3. Rate/Term Refinances 26% PD penalty relative to purchase loan baseline.
PD penalty increased to 30% relative to purchase loan baseline based on new
regression results.
4. Loan Term
15 year loans applied a 40% PD reduction relative to 30 year term
baseline: Loan terms over 30 years were assigned a 100% PD penalty
relative to the 30-year baseline.
15 year loans applied a 50% PD reduction relative to 30 year term baseline: Loan
terms over 30 years were assigned a 150% PD penalty relative to the 30-year
baseline.
5. Investor (Non-Owner-Occupied Properties) 20% PD penalty relative to owner-occupied baseline.
35% PD penalty relative to owner-occupied baseline based on new regression
results.
6. Risk Premium
Risk premium was the eighth ranked PD variable in the regression
model. This variable sought to assess credit risk by comparing the
mortgage note rate relative to prevailing mortgage at loan orgination.
Risk premium was removed from the final PD regression model due to due to multi-
colinearity with other regression variables as well as concerns about its
predictiveness in the current environment. For example, at present, loan mortgage
rates may reflect non-credit issues including mortgage liquidity constraints in
specific segments of the market.
7. Seasoning Treatment for seasoned loans was not included in the exposure draft.
When estimating losses on seasoned mortgage loans, Fitch uses the same PD, LS
and rating stress framework as that used for analyzing newly-originated loans.
However, two key adjustments based on the borrowers updated sLTV and
payment history are made to account for the additional observed performance data
that is available for seasoned loans. Depending on Fitchs view of the borrowers
equity position and the loans performance over time, these adjustments can either
increase or decrease the loans loss expectation compared to the loss level the
model would have been assigned at loan origination.
Loss Severity
1. Quick Sale Adjustments
Quick sale adjustment percentages varied based on rating stress
ranging from 20% in 'AAA' and 10% in 'B'.
Extensive research was conducted into distressed sale discounts over time, and
there was no evidence to suggest that properties sell at a greater discount in higher
stress scenarios. While in higher stresses, more properties will be affected by the
quick-sale, individual property adjustments have remained relatively consistent on
liquidating loans historically at approximately 15% below market value. This haircut
is now applied to all rating categories.
2. Foreclosure Costs - Taxes and Insurance
The agency estimated tax and insurance costs at 2% of the appraised
value, and modelled these as a one-time cost.
The updated approach takes a dynamic view of taxes and insurance, which reflect
costs that will be incurred over the entire liquidation period, or up to 3 years in the
AAA stress. On average, these costs are approximately 2% of the property value
per year based, but are variable by state. Maintenance costs are now assumed to
be a one-time cost equal to 1% of the property value, plus 0.25% per year over the
liquidation period.
Rating Stresses
1. Sustainable Market Value Decline (SMVD) Floors
Assumed 45% at 'AAA' through 20% at 'B' by benchmarking Fitch's peak-
to-trough 35% market value decline expectation for recent correction to a
'A' stress.
The agency now applies a dynamic two-step process whereby properties are first
adjusted to their sustainable values and then applied further sMVD stresses that
correspond to different rating scenarios. This approach better considers where we
are in the housing cycle, with stresses and CE increasing in 'unsustainable bubble
scenarios' and declining in environments where prices are approaching
sustainability.
2. Liquidation Timelines
29 month foreclosure and liquidation timelines for the 'B' through 'A"
rating categories were established based on observations in current
environment and assumed further extended timelines for 'peak' vintages.
'AAA' and 'AA' timelines incorporated additional timelines extensions of 6
and 12 months, respectively.
Foreclosure and liquidation timelines were shortened in the lower rating categories
to reflect Fitch's expectation that timelines will decrease once distressed inventory
is liquidated and servicing procedures and controls are fully implemented. High
investment grade stresses were only reduced slightly.
3. Economic Risk Factor (ERF) Floors
ERF floors based on UFA's 2010 national risk index values. ERF floors
for all investment grade categories were set at the peak level of 4.7
representing 2007 stressed environment.
ERF floors are based on UFAs 2Q2011 National Risk Index Values, which includes
a revised scale. The agency decided to tier the ERF floor stresses throughout the
capital structure to increase stability in high investment grade categories. While the
ERF floor of 2.5 at the 'A' category continues to represent the 2007 economic
environment (equivalent to 4.7 in old UFA scale), 'AAA' through 'AA' floors
represent more stressful economic conditions.
Structured Finance
ALL FITCH CREDIT RATINGS ARE SUBJECT TO CERTAIN LIMITATIONS AND DISCLAIMERS. PLEASE READ THESE
LIMITATIONS AND DISCLAIMERS BY FOLLOWING THIS LINK:
HTTP://FITCHRATINGS.COM/UNDERSTANDINGCREDITRATINGS. IN ADDITION, RATING DEFINITIONS AND THE
TERMS OF USE OF SUCH RATINGS ARE AVAILABLE ON THE AGENCYS PUBLIC WEB SITE AT
WWW.FITCHRATINGS.COM. PUBLISHED RATINGS, CRITERIA, AND METHODOLOGIES ARE AVAILABLE FROM
THIS SITE AT ALL TIMES. FITCHS CODE OF CONDUCT, CONFIDENTIALITY, CONFLICTS OF INTEREST, AFFILIATE
FIREWALL, COMPLIANCE, AND OTHER RELEVANT POLICIES AND PROCEDURES ARE ALSO AVAILABLE FROM
THE CODE OF CONDUCT SECTION OF THIS SITE.
Copyright 2011 by Fitch, Inc., Fitch Ratings Ltd. and its subsidiaries. One State Street Plaza, NY, NY 10004.Telephone:
1-800-753-4824, (212) 908-0500. Fax: (212) 480-4435. Reproduction or retransmission in whole or in part is prohibited except
by permission. All rights reserved. In issuing and maintaining its ratings, Fitch relies on factual information it receives from
issuers and underwriters and from other sources Fitch believes to be credible. Fitch conducts a reasonable investigation of the
factual information relied upon by it in accordance with its ratings methodology, and obtains reasonable verification of that
information from independent sources, to the extent such sources are available for a given security or in a given jurisdiction.
The manner of Fitchs factual investigation and the scope of the third-party verification it obtains will vary depending on the
nature of the rated security and its issuer, the requirements and practices in the jurisdiction in which the rated security is offered
and sold and/or the issuer is located, the availability and nature of relevant public information, access to the management of the
issuer and its advisers, the availability of pre-existing third-party verifications such as audit reports, agreed-upon procedures
letters, appraisals, actuarial reports, engineering reports, legal opinions and other reports provided by third parties, the
availability of independent and competent third-party verification sources with respect to the particular security or in the
particular jurisdiction of the issuer, and a variety of other factors. Users of Fitchs ratings should understand that neither an
enhanced factual investigation nor any third-party verification can ensure that all of the information Fitch relies on in connection
with a rating will be accurate and complete. Ultimately, the issuer and its advisers are responsible for the accuracy of the
information they provide to Fitch and to the market in offering documents and other reports. In issuing its ratings Fitch must rely
on the work of experts, including independent auditors with respect to financial statements and attorneys with respect to legal
and tax matters. Further, ratings are inherently forward-looking and embody assumptions and predictions about future events
that by their nature cannot be verified as facts. As a result, despite any verification of current facts, ratings can be affected by
future events or conditions that were not anticipated at the time a rating was issued or affirmed.
The information in this report is provided as is without any representation or warranty of any kind. A Fitch rating is an opinion
as to the creditworthiness of a security. This opinion is based on established criteria and methodologies that Fitch is
continuously evaluating and updating. Therefore, ratings are the collective work product of Fitch and no individual, or group of
individuals, is solely responsible for a rating. The rating does not address the risk of loss due to risks other than credit risk,
unless such risk is specifically mentioned. Fitch is not engaged in the offer or sale of any security. All Fitch reports have shared
authorship. Individuals identified in a Fitch report were involved in, but are not solely responsible for, the opinions stated therein.
The individuals are named for contact purposes only. A report providing a Fitch rating is neither a prospectus nor a substitute for
the information assembled, verified and presented to investors by the issuer and its agents in connection with the sale of the
securities. Ratings may be changed or withdrawn at anytime for any reason in the sole discretion of Fitch. Fitch does not
provide investment advice of any sort. Ratings are not a recommendation to buy, sell, or hold any security. Ratings do not
comment on the adequacy of market price, the suitability of any security for a particular investor, or the tax-exempt nature or
taxability of payments made in respect to any security. Fitch receives fees from issuers, insurers, guarantors, other obligors,
and underwriters for rating securities. Such fees generally vary from US$1,000 to US$750,000 (or the applicable currency
equivalent) per issue. In certain cases, Fitch will rate all or a number of issues issued by a particular issuer, or insured or
guaranteed by a particular insurer or guarantor, for a single annual fee. Such fees are expected to vary from US$10,000 to
US$1,500,000 (or the applicable currency equivalent). The assignment, publication, or dissemination of a rating by Fitch shall
not constitute a consent by Fitch to use its name as an expert in connection with any registration statement filed under the
United States securities laws, the Financial Services and Markets Act of 2000 of Great Britain, or the securities laws of any
particular jurisdiction. Due to the relative efficiency of electronic publishing and distribution, Fitch research may be available to
electronic subscribers up to three days earlier than to print subscribers.
U.S. Prime RMBS Loan Loss Model Criteria 35
August 15, 2011